Experiments & prototypes

Lab

Lately I’ve been working with AI tools like Claude Code, Codex, Cursor, Figma Make, and Stitch to experiment with vibe coding and design. It’s a fast-moving space, and I’m learning where these tools help, where they fall short, and what they make possible. A few things I’m exploring:

  • Hearth’s gallery of three illustrated helpers: Grocery Restocker, Expense Splitter, and Subscription Pruner.

    Hearth: household help, on your terms

    A fully simulated agentic AI prototype with three customizable household helpers. Set what each can do independently, when it must ask, and what it must never do. Replay the same week under different rules to see what changes and why.

    The accompanying design system supports AI-assisted implementation through shared component contracts, a generated implementation guide, and live examples. Includes automated checks and a native Figma component pilot with editable states and themes.

    Agentic AI · Trust & delegation · Explainability · AI-ready design systems

  • Active Pac-Man gameplay with a ghost’s target marker on the maze and a feed showing each ghost’s role and intent.

    Ghosts with Intent

    A playable Pac-Man that makes AI opponents’ intentions visible. Four ghosts adapt their roles to how you play, choosing to chase, ambush, cut off, or patrol. A live feed describes each ghost’s intent, while an overlay on the maze shows where it’s headed. Compare what the AI says with what it does, then use that understanding to outplay it.

    Multi-agent behavior · Intent visibility · Explainability

  • Digital Twin answering a question about Kshitij’s design philosophy, with suggested questions and a microphone control beside the message field.

    Digital Twin

    An AI chatbot that acts as my interactive digital persona, trained on my product design and leadership experience. A chat interface with voice input and streaming responses.

    Conversational AI · Voice interaction